Towards functional safety compliance of matrix-matrix multiplication for machine learning-based autonomous systems.

Abstract

Autonomous systems execute complex tasks to perceive the environment and take self-aware decisions with limited human interaction. This autonomy is commonly achieved with the support of Machine Learning algorithms that provide the system with the ability to learn from sensor data. The nature of these algorithms, that need to process large data volumes, poses high-performance demands on the underlying hardware. As a result, the embedded critical real-time domain is adopting increasingly powerful processors that combine multi-core processors with accelerators such as GPUs. The resulting hardware and software complexity makes it difficult to demonstrate that the system will run safely and reliably. This paper tackles part of this challenge by introducing best programming practices and diagnostic mechanisms on the matrix-matrix multiplication function, a central element of existing Machine Learning libraries, in accordance to the recommendations of functional safety standards such as IEC 61508 and ISO 26262. We then evaluate a selection of diagnostic techniques applied to the matrix-matrix multiplication software in the context of integration and execution in high performance multi-core devices. As a result, we provide a catalogue of diagnostic mechanisms with varying degrees of diagnostic coverage for matrix-matrix multiplication execution on heterogeneous high-performance architectures, so that safety engineers can choose the best option for their needs. Moreover, we assess the diagnostic mechanisms on one of the ARM R5 cores of a Zynq UltraScale+ Multi-Processor System-on-Chip and on a Intel i7 processor with native code employing vectorization for the sake of performance.